Digital In-Context Experiments

Brand Safety: Pre-Registration

Authors
Affiliations

Hauke Roggenkamp

Institute of Behavioral Science and Technology, University of St. Gallen

Faculty of Economics & Social Sciences, Helmut-Schmidt-University

Johannes Boegershausen

Rotterdam School of Management, Erasmus University

Christian Hildebrand

Institute of Behavioral Science and Technology, University of St. Gallen

Published

Thursday May 9, 2024, 22:13 GMT+2

This study directly relates to brand safety by highlighting how malgorithms that place advertisements in insensitive or inappropriate contexts—such as promoting travel during natural disasters—can significantly impact consumer perception, resulting in brand dilution and damaging the brand’s reputation.

To explore this, we expose participants in the treatment group to a simulated Twitter feed that consists of real tweets describing the severe flooding currently affecting Brazil. Within this feed, we include a fictitious sponsored post by KLM, advertising flights to Brazil with the following copy:

Brazil’s wild beauty calls! Experience nature like never before. Book your breathtaking adventure with KLM.

Figure 1: KLM Ad

The control group is exposed to the same ad in a different context that does not contain any tweets covering the flooding. Instead, the feeds cover other real topics such as a Madonna concert.

You can browse the flood-related feed here and the more general feed here.

Both feeds are interactive DICE feeds that resemble Twitter and cover posts from various users. Both contain 20 posts where the focal brand’s sponsored post is displayed in fifth position.

Method and Measures

We run a 2-cell (flood-context vs. general-context) between-subjects design in which each participant faces the sponsored post.

After participants browse the social media feed, they are redirected to a Qualtrics survey that starts with basic demographic questions. Subsequently, they answer unaided and aided recall questions to indicate whether they remember seeing a KLM ad.

Finally we elicit the participants’ brand attitudes towards KLM before we debrief and redirect them to Prolific.

Primary Analysis

Our primary interest lies in the effect the context has on our three-item brand attitude measure. Hence, we compute an average of these items and compare means using a simple OLS regression (to which we subsequently add potential covariates).

We expect the flood-context to affect the brand attitude towards KLM negatively, which is why we consider a one sided test.

Secondary Analysis

We expect heterogeneous treatment effects, where recall and the (absence of) the dwell time participants allocate to the focal brand’s sponsored ad moderate the treatment: Those participants who have not seen or processed the ad (approximated by a lack of dwell time and failures to recall the ad, respectively), shouldn’t be affected by the brand logo.

Population

We will recruit participants from Prolific who meet the following criteria:

  • Approval Rate >= 99%
  • First Language == ‘English’
  • Location == ‘USA’

Sample Size

As stated above, we expect our treatment effect to be heterogeneous. As a consequence, the sponsored ad can only affect brand evaluations if it cuts through the content clutter (Ordenes et al. 2019).

Hence, we recruit 300 participants in a first pilot. To this end, we create a database containing 500 rows.

Exclusion Criteria

We will only consider complete observations, that is, data from participants who browsed through the feed, answered the qualtrics survey and who were redirected to Prolific with a functional completion code.

Because we gather process data, such as dwell time, we have tools to assess the data quality (Cuskley and Sulik) – at least during the exposure to the social media feed. If these data reveal inattentive participants, for instance, we may exclude them too but label the resulting analyses as exploratory.

Prior Data Collection

We did not collect any data before.

References

Cuskley, Christine, and Justin Sulik. “The Burden for High-Quality Online Data Collection Lies with Researchers, Not Recruitment Platforms.” Perspectives on Psychological Science 0 (0): 17456916241242734. https://doi.org/10.1177/17456916241242734.
Ordenes, Francisco Villarroel, Dhruv Grewal, Stephan Ludwig, Ko De Ruyter, Dominik Mahr, and Martin Wetzels. 2019. “Cutting Through Content Clutter: How Speech and Image Acts Drive Consumer Sharing of Social Media Brand Messages.” Journal of Consumer Research 45 (5): 988–1012. https://doi.org/10.1093/jcr/ucy032.